Friday 12:35–13:05 in Track 3

How to efficiently model learner’s knowledge with recurrent neural networks

Mateusz Otmianowski

Audience level:
Intermediate

Description

During our presentation we will share the results and experiences connected with implementing state-of-the-art techniques for modelling learners knowledge using Recurrent Neural Networks (Deep Knowledge Tracing).

Abstract

Knowledge Tracing (KT) is one of the most important research areas in personalized education nowadays. It allows us to trace learners’ knowledge over time so that we can accurately predict how they will perform in the future. By improving the quality of such models we can better adjust the adaptive learning experience to the needs of particular students. In recent years the idea of using recurrent neural networks for learners knowledge tracing (Deep Knowledge Tracing, DKT) gained a lot of attention, as it has been shown that it generally outperforms traditional methods. During our presentations we will share the results and experiences connected with implementing this method in one of the Pearson personalized learning products. We will focus on challenges that we have encountered during the model development process related to the framework we’ve used (TensorFlow), training performance, experiment tracking and having multiple people working simultaneously on the same model. We’ll also share the results and compare them with the state of the art results from other papers.

Subscribe to Receive PyData Updates

Subscribe